Re: 'python-build-system'

[ludo@gnu.org (Ludovic Courtès)]

Nikita Karetnikov <nikita@karetnikov.org> skribis:

>> Yes, definitely.  But that’s not a problem: these scripts are most
>> likely unused during the build process and afterward.
>
> So, should I package it (without replacing shells)?

No, no need for zsh here, AFAICS.

>>> * guix/build/utils.scm (wrap-program): Assume that 'prog' is an
>>>   absolute filename.  Adjust 'prog-real' and 'prog-tmp' accordingly.
>>>   Also, change 'prog-tmp' to honor command line arguments.
>
>> What was wrong with the previous approach?  I don’t like the “assume
>> it’s an absolute file name” bit, because we don’t know whether the
>> assumption holds.
>
> Well, have a look at this snippet:
>
> +                  (for-each (cut wrap-program <> var)
> +                            files)))
>
> Each 'file' here will have an absolute filename.
>
> I failed to make it work with the previous version.

Can you clarify what didn’t work exactly?  It’s not clear to me.

>> You need double quotes around $@.
>
> Why should I add them?

So that argument expansion does not incur further splitting, in case of
arguments containing white space (info "(bash) Special Parameters").

>> You also need to keep (package-native-search-paths python).  See commit
>> 35ac56b6, which fixed that for Perl & co.
>
> Could you elaborate?  Why is it needed?  Is it necessary to add
> 'python-search-paths'?

If you append (package-native-search-paths python) to the list of search
paths, then PYTHONPATH will automagically be defined appropriately (see
<http://lists.gnu.org/archive/html/bug-guix/2013-03/msg00158.html>.)

If you don’t, then PYTHONPATH will be undefined, unless you explicitly
define it in build/python-build-system.scm, which is not the recommended
option.

So just mimic the change made in the above commit for perl-build-system.

HTH,
Ludo’.